P2.3: Develop Specialist & RAG Agents

[Virgo-Alpha]

This is a major feature ticket that involves creating the next set of specialized "worker" agents and integrating them with Google Cloud Storage for document analysis. We will create two new agents: a FinancialBot and a powerful RAGSearchAgent for analyzing user-uploaded documents.

Implementation Tips

1. FinancialBot - The Financial Analyst:

• Creation: Use the Agent Starter Pack to create a new FinancialBot microservice.
• Logic: This agent will be a "Formatter" similar to the JsonFormattingAgent. Its instruction prompt will be highly specialized, guiding it to find and extract key financial metrics (Revenue, Profit Margin, Burn Rate, etc.) from a block of unstructured text.
• Schema: It will use a Pydantic output_schema to enforce a structured JSON output for the financial data it finds.
• Tools: This agent will have no tools. Its sole job is to parse and structure text given to it.

2. RAGSearchAgent - The Document Expert:
   This agent allows users to "chat with their documents." It uses a Retrieval-Augmented Generation (RAG) pipeline.

• Prerequisite (F2.7 - GCS Integration): First, you must set up a Google Cloud Storage (GCS) bucket. When a user uploads a pitch deck, your Celery task will save this file to the GCS bucket.
• Vertex AI RAG Corpus: In the Google Cloud Console, create a RAG Corpus in Vertex AI. Point this corpus to your GCS bucket. Vertex AI will automatically process and index the documents you upload, making them searchable.
• Agent Creation: Create a new RAGSearchAgent using the Agent Starter Pack.
• The RAG Tool: The agent's only tool will be the VertexAiRagRetrieval tool from the ADK. You will configure this tool with the ID of the RAG Corpus you created.
• Agent Logic: The agent's instruction will be simple: "You are an expert at answering questions based only on the provided documents. Use the VertexAiRagRetrieval tool to find the answer to the user's question."

Acceptance Criteria (Checklist)

• FinancialBot project is created and can be run locally.
• RAGSearchAgent project is created and can be run locally.
• A GCS bucket for document uploads is created.
• A Vertex AI RAG Corpus is created and connected to the GCS bucket.
• The RAGSearchAgent is configured with the correct Corpus ID in its environment.
• Both new agents can be successfully called via an authenticated curl test after deployment.

Unit Tests

• Local Agent Tests: Test each new agent locally within its own directory using make test.
• FinancialBot Test: Write a pytest test that passes a sample block of financial text to the agent and asserts that the returned JSON contains the correct, extracted numerical values.
• RAGSearchAgent Test: Write a test that mocks the VertexAiRagRetrieval tool. The test should call the agent with a question and assert that the mock tool was called with that exact question as the query.